Sander said:
Holy mother of Jebus, wall of text.
Okay, first, ad AI personalities. What you are talking about here is unpredictability, not necessarily personality differences. Unpredictability can simply be achieved through randomly choosing between strategies that are close in expected results. This doesn't necessitate the overly complicated solution you proposed.
Also, you are wrong that unpredictability is a necessity. While obscuring your strategy and not being predictable is one road to success, another is simply playing really well. See also: game theory.
You are obscuring the issue here by constructing elaborate schemes to try to recreate a human mind. A good AI for a game won't require any of that, it merely needs to produce convincing results. Similar results do not require an underlying equivalent approach.
That simply doesn't work however, the results are never convincing and typically repeat, you've then got an easily exploitable AI, and simply randomizing strategies that will solve the issue is a pretty heartless solution because it just eventually creates the same effect, an AI that doesn't actually reason in a mock form of intuition, but rather just picks random numbers regardless of which is more suitable. As I've said, a good AI
should have a personality to be an enjoyable adversary. I think you're mistaking what I'm saying to be an issue of whether or not the AI is simply good at the game or whether or not the AI is a good opponent.
I'm looking for accurate emulation here, not simply convincing simulation, which doesn't provide the necessary variety and depth. The minute differences a more complex system would create are far more important than their size indicates.
Sander said:
You could do that, but I chose that simple example as it was the first to jump to mind. You could look at turn-based versus real-time RPGs as well (Temple of Elemental Evil versus Icewind Dale 2, for instance), or any other game. But comparing RTS and 4X games is truly unfair.
But it is merely an example to support the main point, that the fundamental difference between real-time and turn-based games is one of a nearly analog, constantly changing game environment and a static environment. As I said: the latter is fundamentally simpler, and hence is easier to create a good AI for.
I'm talking about RTS versus TBS, it's the topic of the thread.
Anyhow, I don't believe the simplicity of the environment is as big a factor as you think it is. As the elements of a game devolve as a whole, yes, eventually a TBS game will be in fact easier to program an AI for. But when it comes down to it, as I've been saying with TBS games in general, there's simply more complexity in the systems they adapt which are harder to develop AI for because there's so much to keep track of that an AI has to respond to intelligently.
It's also an issue of time and how much effort the developer has to take. For an RTS you can develop an AI that will pretty much function on all map types, will be able to utilize all the abilities within the game, and simply use them all functionally, because there's just a lot less for the AI to have to process on a fundamental level regarding the game's systems.
That means there's a lot less to do as a whole.
We were talking about the diplomacy system for example, I think it's good that you brought up that they eventually have to develop these systems independent from each other because although these systems are different entities, they eventually effect
all different aspects of the game, the AI has to predict what will happen to his economy if he adopts a new religion that requires higher upkeep but raises the happiness of his people, he has to predict what will happen to his army if he institutes a new government which promotes pacifism and bolsters the economy, he needs to know what will happen if the way he interacts with other plays during diplomacy will hurt certain factors of his empire, if trading a technology will be beneficial or strengthen a possible future enemy he's on tenuous terms with, you were saying that an RTS AI needs to predict things too, but it simply isn't as important in an RTS game, mostly because there the order of the day is react, react, react. I see this, I do this, I build my simple base to get to tech level 5 then make this sort of unit because it is stronger. Simple things, things that in a TBS game aren't so simple, that are usually far more involved and affect hundreds of other variables.
I mean, a human player acts the same way in a RTS game, he does formulate strategies and predict his enemy's movements, but for the most part he is reacting, reacting to occasions in battle or to changes in his economy which can be fixed with a simple action, rather than having to alter fifteen hundred components of his entire government, religion, and army. My point is that reactions have a lot less possibilities to account for, because they're so basic to comprehend. If you're losing, you retreat, you're getting attacked, you send units to help.
There are always more complex situations to be accounted for, but in the end, they're never as advanced as a TBS's issues where the computer has to act and not simply react.
Sander said:
This is also part cause and effect. Designers keep real-time systems relatively simple to avoid these AI problems.
Also, your last large paragraph is actually a pretty good advocate for why real-time AI is harder to code for, as you're simply saying that a real-time AI can only quickly react and not make any long-term plans. Heh.
And since the genre has evolved on that principle it's easier to say that practically it's much easier to develop an AI that functions adequately.
As I said before, reaction is the most important thing in an RTS, and since an AI can do that in an extraordinary amount of time it naturally has an advantage, its inability to plan is a definite issue, but not as major as it would be in a TBS game where that's practically the focus.